Episode Transcript
[00:00:05] Welcome to Prompting Curiosity, a podcast for the AI curious. No coding background required. I'm your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what these AI tools actually are really though, are the files in the computer, how to use them and what they might mean for how we think, work, create and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it.
[00:00:38] Hello, hello, hello, my curious people, and welcome to episode 55, uh, of prompting curiosity. I'm your grateful host, the Maestro, and today we are talking about open source and open weight AI models. We're going to hop right into this episode because I think it's going to be a little bit longer. I'm looking at the outline. I got a lot to say about this. I didn't think I was going to. When I was sitting down and outlining this and thinking about the episode, I was like, oh, how long will it be? And then I typed it all out and I was like, it will be quite long. So we're going to hop on in. So open source, open weight. Maybe you've heard of these terms. I have mentioned them a little bit before, uh, in a previous episode. Uh, previous episodes, especially episode 17. Will chat GPT get old navied. I will link that in the show notes, but I'm going to use the term to start off with open model. Uh, and that is a term and a concept that I want to make sure that you know about because I do think that it has a very future proofing nature to it. And also there are some new models being released that are, they are specifically open weight, um, but they are making news and maybe you've heard of them, namely Kimmy K3 or Deep Seek. Uh, and I want to make sure that my curious people, that's, you know, what is going on. So before I get into what we're going to put into the umbrella term of, of open models, before I get into what they actually are, I want to explain, which I want to zoom out and I want to explain why they are even a thing at all, right? Like why they come up in AI conversations, why you'll. Why you will read about them in AI news and why the companies are, we'll call it public, we'll say publicly fighting about them. And there's more, um, we'll talk more about that at the end of the, the end of the episode. But it really comes down to two things, how these models affect the market and what they can do for the individual user. So remember, an LLM, it's just a computer program, albeit an insanely complex one.
[00:02:42] But LLMs like Claude and Chap, G, B T, they are. What are. They are what is known as a closed model, right? Which means that only Anthropic and OpenAI, respectively, can actually run them, right? We just get to use the model. We don't host them, we don't run them, we just use the model.
[00:03:01] Open weight models, they give the users, and that user could be a company or it could be an individual. They give the user the ability to actually run the LLM themselves.
[00:03:14] All right, so from a market standpoint, open weight models can help keep prices in check. And this is one of the things that's really fascinating to me. There's a privacy aspect that we'll talk about a little bit and we'll talk about a little bit in a little bit. Um, but this, uh, market side of things is what's most interesting to me. So right now we have a handful of companies. Anthropic, OpenAI, Google, etc, they control access to their best models completely. Absolutely right. Those models are closed, meaning that you only get to use them. You, you don't get to do anything else, but you don't host them or anything like that.
[00:03:53] Anthropic, OpenAI, Google, they set the price and they can raise that price whenever they want. And you either get stuck paying it or you don't get to use that model.
[00:04:03] Open models break that, right? Uh, once a model is released, once one of these open models is developed and released, the recipe is out there, they put it out, it's out there permanently.
[00:04:14] And companies, multiple companies at the same time can download and then host and run this model. And then that means they get to compete with each other on price, right? Uh, yes, folks, there are companies that their sole job is to host and run open weight AI models and then they charge per use, right? But because multiple of these hosting companies exist and they can all serve the same exact open weight model at the same time, they compete with each other on price and that mechanism keeps the price down. So if one, you know, company wants to raise it, the other company's like, I have this exact same model and I'm charging a, uh, lower price and I'm going to get more customers. That's good for us, right? That's actually very good for us. On the flip side of this, right, from the individual user standpoint, it's about Sovereignty, right. So again, I'm going over the main reasons why you're going to even hear about open, open models in general.
[00:05:15] We went over, the first point was like, what it does to the market. But the second point is, uh, what it does for the individual. And it's really about sovereignty and not being dependent on one company's decision about access to a tool.
[00:05:28] Right? If a closed model provider, like I was just speaking about, if they change their pricing or they change their rules or they just shut something down, you know, they don't have. You don't have any say, right? So if, if Claude, if Anthropic decides to change the prices, just jack them up to be a million. You can't say anything. You're just either like, well, I guess I won't fucking use that thing, I can't use it, or I'm just going to pay it.
[00:05:49] With an open model, you are never fully at the mercy of one company because the thing itself, the model, right, the thing itself is not owned by a single gatekeeper anymore. That's pretty dope. And if you go to the most extreme, and I say extreme in a good way, if you go to the most extreme, Tevy, you know, tevy, wow. Tech heavy version of this scenario, that would be you running the model entirely on your own machine and having basically full control over how you use it, right? Nobody could change your price, they couldn't shut off access, they can't see your data, they can't take the model away from you. That is like the most sovereign way of using these open models, right?
[00:06:30] So let's zoom in now and get into the definitions and learn a teeny bit about the tech side. Right. So I have been saying primarily for the, for the episode thus far, open model as the term. But what you will often read about or see online are the terms open weight and open source.
[00:06:51] So open weight means that, uh, the developing company takes the model that they trained and they just release it, right? They hand you the finished trained version and you can download that and run it yourself. That's open. Wait, open source. This is like the real strict definition and it's a much bigger ask. This means that the developing company, they give you the weights. So that's the final model. They also give you the training code and the actual training data. But plus they give you enough or they publish enough documentation that someone could rebuild the whole model from scratch if they wanted to. Right? That right there is the actual bar for open source. And almost nothing that's out there actually meets that standard, right?
[00:07:38] Primarily, for the most part, anything that you're hearing about, if they say it's an open model, it is an open weight model, right? That means that the company has published the finished, trained version, and you can just download it and run it on your own. Okay, so we'll get a little bit more nerdy for a second, and I want to dive into why it's called open weight. So for any of my nerds in the crew, this section's for you. If you're like, I don't care, then you can. Y' all can fast forward just a little bit, right? But why is it called an open weight model? So we're going to take it way back to like episode one where I discussed parameters. Now, parameters as it relates to an LLM can be a little bit difficult to conceptualize, but in essence, they are simply the connections between patterns in language. Example, grammar, tone, rhythm, logic, things like that.
[00:08:29] And these patterns get a weight assigned to them, right? We, you know that LLMs AI, uh, it is just math. And that's what we mean when we say a, uh, weight during training the model. Right? This is how these LLMs get trained. They get fed a ton of content, right? Basically everything on the Internet. I'm not saying it's legal, it's mostly very illegal, but this is how it gets trained.
[00:08:52] They get fed everything on the. On the Internet, and then they get tested on that content. The LLM gets tested on that content. And the test, how it works is that the model gets presented parts of a real sentence, such as the sky is blank.
[00:09:07] And then it gets asked what comes next, right? What fills in that blank? The model will then guess, AKA it completes the sentence. And if it's wrong, it uses math to adjust the weights between language patterns, right? The parameters to increase the probability that it will guess correctly the next time. Right? So if it said the sky is black and you're like that. Well, kind of. But what we're looking for there is supposed to be blue. So it's going to do the mathematical computations, right?
[00:09:37] Because it has a list. Let me go back a little bit. When it's going to finish the sentence, it's basically picking from a group of words a list of words, right? So it'll say the sky is blank, and it's going to pull from, like, a list of words based on the probability that's assigned to each of those words and the one with the highest probability. That is what it generates. That's what it puts as the, as the fills in the blank there. If it's wrong and it's like, hey, it assigned a uh, 98% probability that it's correct for the word, you know, the sky is dancing and it's like that is not right. So then it would do math to recalculate and re address and assess and uh, set, reset the, the weights for these connections so that next time when that has a list, what would show as being, you know, 98% likely that to be the correct way to finish the sentence would be the word blue. Um, so the more parameters, AKA the more connections between pattern, all the different types of patterns in language, the more parameters that a model has, the more finely tuned the responses from that LLM can be. So when you hear that this is like a 2 trillion parameter model, that's what they're talking about. And that means that we can get a really fine, fine tuned, um, response.
[00:10:57] This fine tuning occurs during what's called pre training. And then the weights are quote unquote frozen, meaning they stop being tuned. They could be tuned later on if it's an open weight model. But for what we're talking about here, for a closed model, it means that they stop training it, they freeze it in time, quote unquote freeze it in time. And that's what the model they're going to, to launch. Right?
[00:11:18] And they're going to launch it for public, public use. Right.
[00:11:23] So an open weight model gives you those trained weights, right? The finished set of numbers that make the model work and it's just ready to download and run. That is what's being referenced whenever you hear the term open model. Right. Ah, so again, the majority of what's being released in this, you know, quote unquote open category are actually open weight models. Right. Uh, despite the fact that folks are using lazy language and calling them open source.
[00:11:54] The term open A. Wow. The term open source AI has basically been, you know, claimed by marketing at this point.
[00:12:03] So models that you may have heard of, maybe deep Sea, Quinn, Kimmy, they are open weight. They are not open source. Even though it may say open source, they are not truly open source, they are open weight.
[00:12:16] It is worth noting fun fact that genuinely open source models do exist. They're just like in a completely different tier of conversation. Right? And that's more so involved with like academic infrastructure than a consumer product.
[00:12:30] So so true open source models, they are typically built by researchers who want other researchers to study exactly how a model got made. Right? They're not being built by companies trying to win the best Model AGI race. Okay, so let's jump in. Let's, let's change gears, um, and talk about the actual open weight model. I kind of gave you a little bit of a list there of names. So let's use that as a segue and, uh, just quickly go through what models, what open weight models are actually out there right now and who's putting them out. So meta, they do have one. Theirs is called Llama Google, it's called Gemma Alibaba. They made Quen Moonshot. AI has made Kimmy. That's like the big, big, big player right now. Deep Seek, maybe you've heard of. They make a model called Deep Seek. Mistral. AI makes a model called Mistral. Um, there's one, I don't know how you pronounce it. Zifu.
[00:13:23] I don't know. It also goes by Z AI now they make glm. And Open AI does have one and it's GPT us or maybe os. I believe, I believe it's actually oss though. But now that we've covered that, I just want to go through that. Like I said, I want to go to that super quickly.
[00:13:43] Uh, and hopefully you're like, oh, some of the big names in there.
[00:13:48] What, why? And I want to use it as a jumping off point to circle back to the opening discussion of why, right? Why these models even exist. Why do these companies put them out? Why should we not? Should we? But why might we think of using them? So in the meaning of the episode, we discussed the market and sovereignty implications, but hopefully after you see those, those names, right, Google's in there, OpenAI's in there, Meta's in there. You're asking yourself, why would the companies want to make these?
[00:14:16] So before we get into this, I know I'm like leading up and then pulling back and leading up, but two things that are worth highlighting.
[00:14:24] Open weight M does not mean free, but it is typically associated with a lower price.
[00:14:31] Second thing, in order for any of these models to make any kind of a splash to be in, you know, pausing and uh, causing being in the news of any kind, they need to be capable, period. All right, so if an open weight model is making headlines like, like Kimmy is, it's because it has satisfied both of those criteria, right? It's comparable to the frontier models. Claude chatgpt and it is also in some way cheaper.
[00:14:57] So let's start with that, right? Regarding the price here, what it means that, or rather my back it up regarding, uh, pricing, when we're looking at using an Open weight model.
[00:15:12] There tend to be three separate price points depending on how you're accessing the model.
[00:15:18] The most techie way would be to run the model fully locally on your own hardware. And then if you're doing that, uh, you've downloaded it, you're running it on your own computer, there is no per charge use at all, you just have it on your computer. Uh, your only cost is the electricity and then whatever you paid for your, your computer. I'm going to go in in a little bit as ah, to like loosely like the specs for like what you would need and only the smaller models could you run on like a home unit. Otherwise you would need like a big, big, big, very expensive setup.
[00:15:50] So that's one way that you would access an open weight model, um, and the price associated. Right. So if you're downloaded it to your own device, you're running it locally, then there's no charge to use it. Right. Your only cost would be electricity and just whatever you paid for your hardware, if you are accessing it through a hosting company or an API and you're paying per token, then in that case you're paying per token and you're paying the publish rates which tend to stay competitive because of that fact like we talked about earlier, that multiple companies can host the same model and they will choose to undercut each other.
[00:16:28] The last way to use it, uh, one of these open models is the consumer app. That many, and by many I mean most of the models have them. Right. Nearly every open weight model has a free, has an app, um, and it has a free tier and it's often pretty generous. Um, you know, unlimited basic chat in some cases with only the heavier features being gated behind a plan. So as soon as you start to go into like the coding kind of things then, then you know there is a, there is a payment for that.
[00:16:58] But again the real pricing safety net is the fact that because the model itself is open and anyone could run it, multiple companies are hosting it which means they are all competing with each other on um, price and that keeps the price low. Okay, so the second thing here, right I had two, I said two things worth highlighting. Open does not mean free. So we wanted to talk about the pricing. And the second thing worth highlighting is that these models have to be capable so, so as not to you know, bore you with benchmarks that neither of us understand.
[00:17:29] Uh, and not you know, I don't want to make this episode any longer than it needs to be. Um, when it comes to capability, the closed Frontier models, Claude, uh, you know, Chat GPT, they are still at the top of the leaderboard. But the open models, the open weight models are genuinely competitive and that is what's making such a big splash. Kimmy came out and it made some noise and people are like, wow, this thing is capable. And, and in some cases it is cheaper, right? You still, they're still paying for it to use it. Remember, remember, open doesn't mean free.
[00:18:05] Um, but it can keep prices down and it can ultimately provide a solo, you know, an alternative. Should Claude be like, well, we are jacking up the prices and it's like, okay, well now this, this model, Kimi K3 that was, let's say that even the paid tier was comparable to, to Claude. If Claude jumps up, Kimmy stays at that same lower price. Suddenly it is cheaper, right? So again, as it relates to capability, the gap between closed models and the best open weight models has gotten very small and at least small enough for the, you know, for the vast majority of tasks that, you know, regular, regular users like you and me and businesses are actually needing it for. That gap barely matters anymore, right? And uh, uh, there are a few specific areas that I was reading about that these open models are actually winning. All right? And so I'm very, very, very interested, very interested in this, right?
[00:18:59] So what follows logically and what I was segueing into from before is, you know, you see these big names, uh, you know, Meta and uh, Google and OpenAI having these open weight models and it's like, why the fuck, why would they develop them? Well, there's a few reasons why any company would develop them, right? Infiltration. If everyone builds their tools and their startups on top of, you know, Llama, let's say that's Meta's, um, open source model, that Metal Meta doesn't need to charge it, right? They have made them themselves the foundation for, for things and everyone else depends on that and that is beneficial for them. Um, it's free R and D. When you make weights public, thousands of developers start using it. Sorry, whistle there, start. Thousands of developers start using it. But you know, poking at it, breaking it and you get feedback, uh, without having to pay anything.
[00:19:58] Uh, I've seen it written up as a recruiting move. So the best AI researchers want to work, you know, they want their work seen and used, not buried in a closed product. So companies that publish their, their models, they get first pick or, you know, better pick of the talent. I'm sure they, you know, also they want to make the most money but that is one thing, uh, undercutting the market. So this isn't necessarily why a big player would use it, but this is why other people and, and would come out with this model at all, is to undercut the market, right? If you're not in the, you know, if you're not the best model, if you're not in first place, okay, this is like companies like Deep Seek Moonshot with Kimmy. You can undercut, simply just undercut the whole game by giving your version away for free, right? You steal attention and you get people into your ecosystem and then they can make money because they, uh, do have paid tiers. I.
[00:20:50] Lastly, for some of these bigger companies, the model itself is not the business, right? We know meta Google. Like they're not making money off of AI. It's not their main play. So a free model just gets more people, you know, building on their infrastructure and into their ecosystem. And that's the actual payoff, right? The model is just the hook, it's just the thing that brings them in.
[00:21:11] Um, flip side of this, and hopefully I've articulated this or you've at least been able to deduce this from what I've been saying is why would an individual want to use one of these, right? So we just talked about why a company would release this, but why would you listening to this, perhaps choose one is cost, so perhaps it's not, you know, it's not a big deal right now. But if these models do decide to jack up the price, then suddenly, uh, don't be surprised if you see more people looking, right?
[00:21:35] So we know this. When multiple companies can host that same open model, they can, they compete on price. So you're not ever stuck with just one company. And if it decides, you know, they want to jack it up, you're like, well, that's fine. I can use the same model that's being provided by a different company.
[00:21:50] Similarly, you're not locked in, right? If a provider jacks up the price or disappears, you can literally switch to a different host that's running the same model.
[00:22:01] If Claude does that, you can't use cloud anymore. If ChatGPT does that, you can't use ChatGPT anymore. You're, you're, you're out of luck.
[00:22:09] Next one. This requires more tech here, but if you have privacy concerns, whether, whether it's sensitive client work, and I'm thinking about a client, a friend right now, you have sensitive client work, medical stuff, legal stuff that you don't want to go onto, you know, company servers, you Run the model locally on your own machine and nothing ever leaves your computer. Um, and lastly this again ties into running things locally and it's more tech heavy. But this is true customization. So you can fine tune. Remember I said earlier that the weights are frozen, quote unquote frozen, but that just means that they stopped training them. When you have an open source model and you're running it locally, you can actually continue to fine tune those weights, um, whether it's on your own data, your voice, your specific use cases, whatever, and really customize this model. But again it has to be uh, you know, while you're, you're using this, um, running this locally or you have access to that whether you're running it locally or hosting it on, on rented infrastructure. Um, so it's again it's a more techy, techy option, uh, there. Right. But moving on, for those of you that are like, you know, maybe, maybe I'll give it a try, right, You're a little curious and you want to try one of these open weight models, you have a few options. One, just go to the free app or you know, their website. This is the simplest option. There's no download, there's no setup. It works exactly like Claude does. Exactly. ChatGPT.
[00:23:31] Most of the major open source models have one. Kimmy has its own, Deepseek has its own, Google has its own through AI studio. They all have it is what I'm trying to say. So easiest places to start there. So if you're like, hey, I'm here in Kimmy, just literally type in Kimi M.com or just Google Kimmy and it'll take you to the site. And you, you do have to like put in your, in an email, um, like all of them do. But uh, that's, that's it.
[00:23:58] Uh, the second option for trying one of these open weight models would be what's called a hosted quote unquote playground or an API. Still no download required. Um, if you are a bit more tech savvy, you want more control, you're a developer building something, you can access the model through the company's API instead of the, the consumer app.
[00:24:18] Again, there's still no download of any downloading of anything. You're using the company servers, you're just accessing it through a different door. And you will typically in this case be billed per use instead of a flat monthly price. Right, Claude, ChatGPT, they have the same type of thing, the API access as well.
[00:24:36] You could also go through a third party host instead of the Original company.
[00:24:40] Um, there's services like Open Router together AI or Grok, but this, the Groq, um, they host, they host these open models and you can access them that way and sometimes they are cheaper uh, than going directly to the company that built it. Right.
[00:24:57] Lastly we have the most tech heavy uh, way of accessing it which would be to download it and run it yourself. Right. This is the fully local option. So you get the actual weights file from something like Hugging Face. I know these are the weirdest fucking names but that is, it is a thing. Uh, you go website called, it's a, like a repository but you go to a website called Hugging Face and um, then you use a, you can use a free tool like Olama or LM Studio to actually download and run the model on your computer. Again like I said earlier, this only works for the smaller models. The largest ones like Kimi K3 that I've been talking about, they require far, far, far, far more computing power than a personal computer has. So you'd be looking at different, very much smaller models that are still very capable. Um, but you're not going to get like an enterprise level model running on your, on your Mac studio at home. Right.
[00:25:51] But uh, I will probably give Kimmy a try at some point. Um, but, and I'll do it through the, the app. That's how I know they have just the regular desktop um, web app.
[00:26:02] Uh, because I was looking at it to make this episode but I really wanted more than anything I wanted to make this episode to make us all aware, myself included of, of what, what you know, an open weight model actually is. Because I, I really do think that they have a ton of value and not in terms of like oh it's going to make us money. But I do think there's a huge future proofing component there. And fun fact, I am not the only one who thinks this. So actually on July 24th, so just like a, a week maybe before, what is that today the 31st and I'm recording this. So a few days ago July 24th, um, Nvidia's CEO Jensen Huang, he publicly posted a letter. It was like his first post on, on X.
[00:26:48] Um, but he posted a letter co signed by 25 companies including Microsoft Meta Hugging Face. And the letter was arguing that America's, the United States's AI advantage depends on a broader open ecosystem, not just you know, these closed models. In this, you know, one company's best model. This letter came in response to two things. One, accusations from the White House insert massive Eye roll that. Moonshoot. Uh, Moonshot, which is a Chinese company. They, they, um, what's called Distilled. They distilled. They used another model. They used Fable 5 in order to build their own model. That is what the White House is claiming. It's the first time ever that like a company has been called out by name.
[00:27:33] Um, so that's one thing. And then the second thing is just that we're. There's currently a bigger fight in Washington as to whether open models should be restricted outright.
[00:27:43] Interesting. Anthropic did not sign this letter. Right. Their CEO did eventually put out a statement saying that, you know, they're not against open models as a category. They just want safety testing required across the board whether it's open or closed.
[00:27:58] Um, but the letter kept growing and the signatory list, it was at like 50 people in a. In a single day. Big names. Right. Including OpenAI and Google, though they joined a little bit, but, um, afterwards. But they signed on board as well. Right.
[00:28:13] So, you know, it's not lost on me that the letter was written by the CEO of Nvidia, which is a company that stands to benefit tremendously from anything AI related. Like, literally anything has to do with AI, they're going to benefit. They make chips, right?
[00:28:30] So open or closed, they are going to benefit. So it's not lost to me. They're like, yes, more AI, more better. But I have expressed my concerns numerous times about the fact that companies like Anthropic and OpenAI have us by the proverbial short hairs. And I do believe that open weight models are the solution.
[00:28:51] All right, last things last. How I. How I wrapped it up. Wow. How I used AI this week. And then we will wrap it up. So if, uh, you're new here, welcome. Each episode I share a quick example of how I used AI that week. So this week I use a new cloud feature that I want to put you onto. It is called scheduled Tasks. So Chat GPT has had this for a minute, and by a minute, I mean a long time. I'm, um, going to actually talked about this in a previous episode and it was one of the, um, how I used AI this week sections. Um, and I used it to search for deals for an outdoor camera and also for an ssd because I was looking for them to go on sale because since AI is a thing, SSDs have gone through the roof the price of them. Um, but I had it as like a search that was set and it ran every Monday and then sent me an email, um, but this time, I set up a scheduled task in Claude that will run every Wednesday morning. And it delivers me a prompting, curiosity, brief. So when I sit down to write these, to write, to record these episodes, but I go to outline them first. I always chat with Claude about topics especially, I'm not gonna lie, like, things aren't changing that much in the AI space. And so I'm like 55 episodes in, and I'm like, what am I talking about today? What am I gonna talk about? What would be interesting? What would be helpful for my people? And so I have conversation with Claude. So I decided to just automate that. That process there, that process, if you're in Canada, um, and have Claude deliver me some topic ideas every Monday morning, along with. And this is actually what I'm more excited about, along with two ideas for me of things for me to try with Claude. So one suggestion that I wanted to give me is a feature or capability in Claude that I could experiment with. And a second thing I wanted to suggest is something I could build, uh, to automate a task in my business.
[00:30:31] Right? So that's one of the things that I see as one of the, you know, a.
[00:30:36] An omission, a hole in. In when it comes to AI. Is that, like I said a million times, AI is very much, here's a problem, go find a solution.
[00:30:45] Uh, so I wanted to just offer me up a bunch of, you know, excuse me, I said that backwards. Here's a solution, go find a problem.
[00:30:54] Um, so I want it to offer me up some stuff, some ideas, some use cases, um, based on what it knows about the business. And so it's like identifying a problem, giving me solution, and then I can decide if I want to do that. Um, but it has actually run one time already.
[00:31:08] Um, but I already had this topic in mind, so I didn't need the suggestions that it offered. Uh, but I did like one of the builds that it suggested. Uh, however, I am busy, as right now with work, so, um, I don't have time to be tinkering around, but I do have it saved and I will give it a shot on a rainy day. All right, I'm looking at the time. This is a longer episode, but I told you that in the beginning. I did prepare you.
[00:31:30] Uh, and that is all for today. Hopefully you found this episode helpful. If you did consider my friends leaving a little rating or A little review, 5 stars would love it. Your voice, your words, your thoughts. Would love to read them. Don't forget, I also have a companion. Wow. A companion newsletter and blog called the Curious Companion, and that drops every Thursday. And that is basically by basically I mean exactly the podcast episode in text format. So if you prefer to read or you just want a written record of the thing, you can join the newsletter fam or on over to the blog to check them out. Head to prompting curiosity.com forward slash newsletter or prompting curiosity.com vlog. Or keep it simple and just check out the link in the show notes. I gotcha. As always, endlessly, endlessly, endlessly appreciative for every single one of you. Until we chat again next Thursday.
[00:32:25] Stay curious.
[00:32:29] It.
